DeepSeek's New Model Claims Sharp Cut in HBM Usage; Samsung and SK Hynix Plunge Over 3% Intraday - finance.biggo.com

Chinese AI startup DeepSeek unveiled its V4.1 Flash model, claiming a dramatic reduction in High Bandwidth Memory (HBM) usage, sparking concerns over declining semiconductor demand. The model reported... Read original →

DeepSeek's new model significantly reduces HBM consumption, directly pressuring the valuations of South Korean memory chipmakers. If its KV Cache optimization becomes widely adopted, demand for high-end memory chips could decline, explaining the intraday drop in Samsung and SK Hynix shares. However, this advancement may not undermine overall AI compute growth, especially in training scenarios still reliant on high-bandwidth memory. U.S. players like Micron show stronger resilience due to diversified portfolios. From a supply chain perspective, if Taiwan's foundry capacity shifts toward AI chips, global memory overcapacity may intensify. Short-term market sentiment is pressured by Fed rate hike speculation, increasing tech stock volatility. In the long run, AI efficiency gains may drive compute democratization, but memory firms must pivot toward high-value storage solutions to avoid structural overcapacity.